摘要
针对帧到帧模型里程计中特征点的冗余、耗费计算资源的问题,提出一种自适应特征提取与匹配的视觉里程计算法。对局部地图特征区域分块,在已有特征区域分块的基础上,基于稀疏化保留冗余特征区域中的高效特征点;针对特征区域中特征匹配点不足的情形,从局部地图中补充方向性和尺度性良好的ORB(oriented FAST and rotated BRIEF)特征点,利用补充的特征点二次匹配;结合PnP(perspective-n-point)估计姿态,实现SLAM的前端视觉里程计。采用TUM(Technische Universit t München)通用数据集验证,并与其它算法在前端时间、特征点数量、轨迹绝对误差等方面对比,对比结果表明了改进算法在上述特征效果的优势。
Aiming at the feature point redundancy and the computing resources consumption in visual odometer,an adaptive feature extraction and matching method for visual odometer calculation was proposed. The local map feature area was partitioned,and the high-efficiency feature points in the redundant feature area were reserved based on the sparse information in each feature area. Considering the lack of feature matching points in feature area,ORB (oriented FAST and rotated BRIEF) feature points with good directionality and scale were supplemented from the local map and they were used for second matching. The front-end visual odometer of SLAM was realized by estimating pose with PnP (perspective-n-point). TUM (Technische Universit t München) dataset was used to verify the proposed algorithm. Compared with other algorithms in front-end time,the number of feature points and absolute error of trajectory,the results show the advantages of the improved algorithm in the above-mentioned feature effects.
作者
徐彬彬
刘鹏远
张峻宁
XU Bin-bin;LIU Peng-yuan;ZHANG Jun-ning(Missile Engineering Department,Army Engineering University,Shijiazhuang 050003,China)
出处
《计算机工程与设计》
北大核心
2019年第7期2076-2081,共6页
Computer Engineering and Design
关键词
机器视觉
视觉里程计
局部地图点
特征区域稀疏
特征区域扩张
特征匹配
machine vision
visual odometry
local map point
feature region sparsity
feature region expansion
feature matching
